AI TrackerFrontier desk · AEST
Back to news
normalaitech

CLIFT: Conformal Self-Verification for Web Agent Training and Test-Time Scaling

arXiv cs.AI · 6 Oct 2026, 03:57 AEST

CLIFT introduces a method for training web agents using conformal self-verification to improve performance. This approach allows agents to utilize their own verification processes during both training and testing phases.

Key facts

Read aloud

CLIFT introduces a method for training web agents using conformal self-verification to improve performance. This approach allows agents to utilize their own verification processes during both training and testing phases. Open-source web agents are now strong enough to execute realistic browser tasks, but training them with reinforcement learning still depends on weak supervision. During training, the agent answers natural-language verification questions about its own rollouts; a Compositional Conformal Certifier keeps only question signals whose URL-conditional evidence agrees with a training-time judge. At test time, the same certified bank is frozen and reused as structured evidence for Conformal Trajectory Selection CTS. On WebArena Infinity, CLIFT achieves state-of-the-art performance among open-source web agents. Together these results position conformal self-verification as a way to turn costly judge feedback into a reusable training signal and a judge-free test-time scaling signal.

Sources in this cluster